Approximate Numerical Expressions (ANEs) are linguistic expressions involving numerical values and referring to imprecise ranges of values, illustrated by examples such as " about 100 ". In this paper, a general principle is proposed to interpret uncontextualised ANEs as intervals of denoted values. It is based on an empirically justified combination of characteristics of numerical values, both arithmetical and cognitive, and in particular, taking into account the cognitive salience of numbers. This general principle is instantiated in two computational models that can be extended so as to take into account the applicative context. An empirical study is conducted to assess the performances of the two models, comparing them to state-of-the-art methods, on real interpretations collected through an on-line questionnaire. Results validate the proposed characteristics used to build the models and show that they offer the best performances in estimating the median interval chosen as representative of the collected intervals.